AI traffic is growing, but our measurement isn’t keeping up.

In summary

• What: AI platforms are increasingly influencing how consumers discover brands, but much of that activity remains difficult to identify in website analytics.
• Why it matters: AI referrals can be lost or misclassified, while traffic from Google AI Overviews and AI Mode sits within Organic Search. The AI traffic marketers see in their dashboards may only be part of the picture.
• The challenge: Better tracking can improve the data we collect, but it cannot recover referral information that was never passed in the first place.
• What marketers should do: Improve how AI traffic is classified, investigate gaps in existing analytics and combine multiple measurement approaches to understand AI’s influence on customer behaviour.

AI traffic isn’t as easy to measure as it looks

AI traffic is starting to appear in marketing reports, with platforms such as ChatGPT, Gemini and Claude increasingly directing users to websites, while Google Analytics is getting better at identifying these visits, giving marketers another source of traffic to analyse alongside Organic Search, Paid Search, Social and Email.

But how much of that traffic are we actually capturing?

Testing published by AI and SEO consultancy WISLR found that GA identified just 20-43% of the LLM-referred traffic generated during its tests across ChatGPT, Gemini and Claude. That suggests actual referral traffic could be 2.5 to five times higher than reported in those scenarios.

It’s a significant difference, but the extent of underreporting will vary depending on each website’s audience, tracking setup and customer journeys.

The research does, however, highlight a measurement problem that becomes more complicated as AI plays a greater role in how people research products, compare services and discover brands.

Some AI traffic is identifiable, some are classified elsewhere in analytics and some arrive without enough information to determine where it came from.

For marketers trying to understand AI’s contribution to website traffic and conversions, the numbers are far from complete.

Mobile makes an already difficult problem harder

One of the biggest challenges is what happens when someone moves between an AI app and a website.

A consumer might ask ChatGPT for a product recommendation, tap a link inside the app and open the website through an in-app browser. They might then switch to Safari or Chrome, return later through Google or complete their purchase on another device.

Each transition creates another opportunity for the original referral information to disappear.
WISLR’s testing found considerable differences between how ChatGPT, Gemini and Claude pass referral information across mobile and desktop environments.

In its tests, some mobile AI app visits arrived without identifiable referral information, making them appear as Direct traffic. Other journeys retained enough information to identify the originating AI platform.

That creates a problem when marketers compare the performance of different AI assistants.
If ChatGPT appears to be generating substantially more referral traffic than Gemini, is that because more people are discovering the brand through ChatGPT, or because its referrals are easier to identify?

Without understanding the underlying measurement differences, it’s difficult to know.

And even when an AI referral is successfully identified, that information may not survive long enough to connect the original interaction with a conversion several sessions later.

Then there’s Google

Google creates a different measurement problem.

Clicks from AI Overviews and AI Mode are included within Organic Search in GA.

A consumer could discover your business through an AI-generated answer, click through to your website and appear in your acquisition reporting as an Organic Search visitor.

That means marketers looking exclusively at their AI referral reports are missing another source of AI-influenced customer journeys.

It also complicates how businesses interpret changes in their Organic Search performance.
If more people are using Google’s AI experiences to research products and services, traditional organic traffic reports alone won’t tell marketers how much of that activity was influenced by AI-generated responses.

Google Search Console provides additional information about search performance, but it doesn’t offer a complete, separate breakdown of AI Overview and AI Mode traffic.

As AI becomes more deeply integrated into search, understanding the distinction between traditional search and AI-assisted discovery will become increasingly important.

Direct traffic isn’t suddenly an AI metric

If AI referrals can disappear, it’s tempting to look at unexplained growth in Direct traffic and assume we’ve found the missing ChatGPT users. Unfortunately, it isn’t that simple.

Direct has always contained traffic that analytics couldn’t confidently attribute elsewhere. Untagged campaigns, messaging apps, redirects, consent settings, browser restrictions and genuine direct visits can all contribute to it.

A business seeing Direct traffic increase while Organic Search declines may have something worth investigating, particularly if the change is concentrated around content that people are likely to discover through AI.

But that pattern alone doesn’t establish that AI is responsible.

Changes in search behaviour, campaign activity, tracking implementations and privacy settings can all produce similar results.

At Louder, we’re interested in investigating these patterns across client data, looking at how acquisition trends, landing-page behaviour, device usage and engagement change as AI-assisted discovery grows.

The challenge is separating genuine changes in consumer behaviour from changes in how that behaviour is measured.

We need to be careful that in trying to solve one attribution problem, we don’t create another.

Better classification can help

There are opportunities to improve the quality of AI referral reporting without pretending every interaction can be identified.

At Louder, we’re investigating how AI-originated traffic is classified in GA and what additional information may be available through our measurement infrastructure.

This includes examining user-agent information in Cloud Run logs and exploring whether signals available through server-side tagging can help identify and classify traffic more accurately.

We’re also investigating instances where identifiable AI traffic may be assigned to unexpected channels rather than appearing in AI referral reporting.

Server-side tagging gives businesses greater control over how available measurement signals are collected and processed before being sent to analytics platforms.

Combined with better referral classification and deeper analysis in BigQuery, it can help identify inconsistencies that might otherwise be missed in standard acquisition reports.
But there are limits.

If an AI app doesn’t pass any useful referral information, server-side tagging cannot reconstruct the original source simply because the data is being collected on a server.

Likewise, user-agent information can help identify certain automated AI activity, but it doesn’t necessarily identify a human visitor arriving after interacting with an AI assistant.

Understanding those distinctions is essential if marketers want to improve reporting without overstating what their data can tell them.

AI visibility is only part of the picture

We’ve previously written about the challenges of measuring AI visibility, particularly how differences in prompts, models and methodologies can produce very different results for the same brand.

Research from SparkToro has since reinforced just how variable AI recommendations can be. Its testing found that AI platforms frequently return different brand recommendations when asked the same question repeatedly, while the prompts real consumers use can vary considerably.

That makes individual AI rankings difficult to interpret as a reliable measure of brand performance.

Referral measurement presents a different problem.

With AI visibility, marketers are trying to understand what happens inside an AI platform. With referral measurement, they’re trying to understand what happens when someone leaves that platform and arrives on their website.

Neither provides a complete picture on its own.

A brand might appear frequently in AI-generated recommendations but receive relatively little identifiable referral traffic. Another might receive more measurable referrals without necessarily having greater visibility across relevant consumer queries.

Changes in branded search, engagement, customer journeys and conversions can provide additional context, but connecting those signals to specific AI interactions remains difficult.

For marketers, the opportunity is to build a more complete view of AI’s contribution rather than relying on one visibility score or acquisition metric.

We’re probably asking the wrong question

Marketing has spent years trying to assign customer behaviour neatly back to individual channels.

AI makes that harder because it increasingly sits across different stages of the customer journey.

Someone could ask ChatGPT which product to buy on Monday, search for the brand on Google on Wednesday and purchase directly from the website on Friday.

The final acquisition report might identify Organic Search or Direct. Neither necessarily reflects the role AI played in introducing the brand or influencing the purchase decision.

That doesn’t make attribution data useless. It means marketers need to understand its limitations.

Rather than focusing exclusively on how many conversions came from AI, businesses should also be looking at how AI-assisted discovery might be changing search behaviour, website engagement and customer acquisition.

Combining referral reporting with AI visibility research, first-party data and broader measurement approaches can help build that picture.

Where sufficient data is available, experimentation and causal analysis may also help businesses investigate whether changes in their AI visibility or discovery strategies are contributing to meaningful commercial outcomes.

The objective isn’t perfect attribution. It’s having enough reliable evidence to make better marketing decisions.

What Louder is doing to close the AI measurement gap

Reliable data on AI visibility is difficult to come by, but as more consumers turn to AI assistants for research and recommendations, marketers need better ways to understand where their brands appear and how that visibility changes.

At Louder, we’re developing tools and methodologies to measure our clients’ AI visibility against their competitors.

Rather than relying on a single visibility score, we’re examining which types of search intent are more likely to generate brand mentions, identifying gaps in visibility compared to competitors and investigating how consistently brands appear across AI-generated responses.

We’re looking beyond simplistic AI rankings towards more meaningful intent clusters that reflect how consumers research products and services.

Variability is an important part of that work. AI responses can change between prompts, models and repeated queries, so we’re monitoring consistency alongside visibility rather than assuming every change in ranking represents a meaningful shift in performance.

Alongside this, we’re investigating how AI referral traffic is classified in analytics and whether additional signals available through server-side measurement can improve reporting.

Our research is still underway, but the early findings are promising. We’re identifying opportunities to give marketers a more useful view of their AI visibility while improving our understanding of the measurement gaps that remain.

We’ll be sharing insights from our AI visibility research in an upcoming article, including what we’re learning about intent, brand mentions and competitor performance.

Subscribe to Louder’s newsletter to receive the findings when they’re published.

Louder’s recommendations

• Treat reported AI traffic as a starting point. Identifiable AI referrals are useful, but they shouldn’t be interpreted as the total amount of AI-influenced traffic reaching your website.
• Audit how AI traffic is classified. Review your GA acquisition reporting, referral sources and channel definitions. Look for opportunities to improve classification where identifiable AI traffic is being assigned to other channels.
• Investigate changes across the customer journey. Analyse Direct and Organic Search alongside landing-page performance, device usage, engagement and conversion behaviour. Don’t automatically attribute unexplained changes to AI.
• Improve the signals you can control. Server-side tagging, first-party identifiers where appropriate, better referral classification and BigQuery analysis can strengthen measurement, although they cannot recover information that was never available.
• Look beyond individual AI visibility scores. Measure visibility across relevant intent clusters, monitor the consistency of AI-generated responses and combine those findings with referral and business performance data.

Get in touch

Get in touch with Louder to discuss how we can help improve your AI traffic measurement, strengthen your analytics foundations and understand how AI-assisted discovery is influencing your business.